--- title: "Apertus 70B API hosted in Europe" description: "Run Apertus 1.5 70B, the fully open Swiss model from EPFL and ETH Zurich, on inference.eu with an OpenAI-compatible API and EU data processing." canonical: "https://inference.eu/models/apertus-70b/" language: "en" --- Chat · Swiss AI Initiative (EPFL, ETH Zurich, CSCS) # Run Apertus 70B in Europe through an OpenAI-compatible API Apertus 1.5 70B is the flagship of Switzerland's fully open language model, built by EPFL, ETH Zurich and the Swiss National Supercomputing Centre. Weights, training data and recipes are all public. On inference.eu you call it with the OpenAI or Anthropic request format, with requests processed in EU data centres and never used for training. [Contact sales](https://inference.eu/contact/)[Example request](#example) 1. 0170B 2. 02EU data centres 3. 03Apache 2.0 ## What Apertus 1.5 70B is Apertus is the open language model of the Swiss AI Initiative, a partnership between EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS). Version 1.5 of the 70-billion-parameter model was released on 24 July 2026 under the Apache 2.0 licence, after training on the Alps supercomputer in Lugano. Version 1.5 continues training the first Apertus with 2 trillion extra tokens of text and multimodal data, for a total of about 17 trillion tokens. It adds image understanding, experimental audio input, an optional thinking mode and a context window of 262,144 tokens, four times that of Apertus 1.0. Unusually, the training data, code and intermediate checkpoints are published alongside the weights. ## What it does well, and where it falls short Apertus is built for transparency and language breadth. Its makers say it was trained only on publicly available data, respecting robots.txt opt-outs, and that it covers well over a thousand languages. That makes it a natural choice for public bodies, universities and companies that must explain where their model comes from. On raw benchmark strength it is not the leader: large open models such as Qwen3, GLM-5.3 or DeepSeek V4 are stronger at hard reasoning and coding. Audio input is still described as experimental. Choose Apertus when openness, documentation and multilingual coverage matter more than the last few points of benchmark performance. ## Calling Apertus 70B through the API inference.eu speaks the OpenAI request format, so the OpenAI SDKs, LangChain, LlamaIndex and the Vercel AI SDK work as they are. Set the base URL to the inference.eu API, use your inference.eu key and set the model to `apertus-70b`. Images go in as `image_url` content parts next to your text. The thinking mode is a chat-template option; use it for maths and multi-step problems and leave it off for ordinary chat. Streaming and tool calls work as usual, and the Anthropic Messages format is also available, so tools built for Claude can switch with a new base URL, key and model name. ## A Swiss model, processed in the EU Apertus comes from Swiss public research, and its makers publish documentation under the EU AI Act's code of practice for general-purpose models. On inference.eu your prompts, images and responses are processed in data centres in the European Union and are never used to train any model. Every customer gets a data processing agreement under Article 28 of the GDPR and our subprocessor list, so you can show exactly where data goes. Billing is in euros. Because Apache 2.0 allows commercial use without user limits, and the training data is public, Apertus is one of the easiest models to defend in a procurement or compliance review. ## Apertus 1.5 70B Maker Swiss AI Initiative (EPFL, ETH Zurich, CSCS) Made in Switzerland Type Chat Parameters 70B Context window 262,144 tokens Licence [Apache 2.0](https://huggingface.co/swiss-ai/Apertus-v1.5-70B/blob/main/LICENSE.txt) Released July 2026 Model ID apertus-70b Processing EU data centres Training on your data Never Source [huggingface.co](https://huggingface.co/swiss-ai/Apertus-v1.5-70B) apertus-70b · Python ``` import os from openai import OpenAI client = OpenAI(base_url="https://api.inference.eu/v1", api_key=os.environ["INFERENCE_EU_API_KEY"]) response = client.chat.completions.create( model="apertus-70b", messages=[{"role": "user", "content": "Summarise this contract in five bullet points."}], ) print(response.choices[0].message.content) ``` ## What teams use it for - 1 ### Public-sector assistants Answer questions from citizens with a model whose training data and methods are fully documented and open to inspection. - 2 ### Multilingual services in small languages Serve users in languages and dialects that most models handle poorly, including many smaller European languages. - 3 ### Research and education Teach and study language models with weights, data and intermediate checkpoints that are all openly available. ## Apertus 1.5 70B: common questions The questions buyers, developers and legal teams ask us most. Something missing? Ask us directly. Compliance depends on how you process data. On inference.eu, Apertus requests are processed in EU data centres, are not used for training and are covered by a data processing agreement under Article 28 GDPR with a published subprocessor list. Apertus's makers also state that training data respected robots.txt opt-outs. Apertus 1.5 supports 262,144 tokens of context, four times the window of Apertus 1.0. That is enough for long reports, contract bundles or entire books. Images take up part of the context when you send them. Yes. Apertus 1.5 is released under the Apache 2.0 licence, which allows research and commercial use, modification and redistribution without fees or user limits. You do not need an agreement with EPFL or ETH Zurich. Most open models publish only weights. Apertus also publishes its training data sources, training code, recipes and intermediate checkpoints, so anyone can inspect how it was built. Its makers describe it as open in weights, data and training details. Apertus was trained on a very wide multilingual corpus; its makers state coverage of more than a thousand languages. Quality varies between languages, so test the ones you need with real content before going live. Apertus 1.5 accepts images alongside text and produces text. It can also take spoken audio as input, but its makers describe audio as experimental. For production speech transcription, use a dedicated model such as Whisper Large V3 Turbo or Voxtral. Yes. Version 1.5 adds an optional thinking mode, switched on through the chat template, in which the model reasons before answering. It helps on maths and multi-step problems. For ordinary chat, leave it off to keep answers short. Both are European public-research models. Apertus 70B is larger, accepts images, has a 256K context and a thinking mode. EuroLLM 22B is smaller, text-only, with 32K context, and focuses on the 24 EU languages. Apertus is the more capable general model. Change the base URL of your OpenAI client to the inference.eu API, use your inference.eu key and set the model to \`apertus-70b\`. Your messages, images and streaming code stay the same. Yes. Apertus 1.5 also comes in an 8B size with the same context window and image support, and the team has released a set of compact Apertus Mini models. The 70B model is the most capable of the family. [Ask a question](https://inference.eu/contact/) ## Related [All models](https://inference.eu/models/#directory) - [EuroLLM 22B InstructRun EuroLLM 22B Instruct, the European open model trained for all 24 official EU languages, on inference.eu with an OpenAI-compatible API.Chat](https://inference.eu/models/eurollm-22b-instruct/) - [Teuken 7B InstructRun Teuken 7B Instruct, the German open model trained on all 24 official EU languages, on inference.eu through an OpenAI-compatible API.Chat](https://inference.eu/models/teuken-7b-instruct/) - [Llama 3.3 70B InstructRun Meta's Llama 3.3 70B Instruct on inference.eu: an OpenAI-compatible API with requests processed in EU data centres and billed in euros.Chat](https://inference.eu/models/llama-3-3-70b-instruct/) - [Mistral Small 4 119BRun Mistral Small 4 119B from French lab Mistral AI on inference.eu: reasoning, vision and code in one model, processed in EU data centres.Chat](https://inference.eu/models/mistral-small-4-119b/) ## Move your AI workloads to EU infrastructure Tell us which models you use and roughly how much traffic you have. We reply within one working day with pricing and next steps. - Reply within one working day - Pricing for your volume - DPA ready to sign